Note
Go to the end to download the full example code
Create a Custom Suite#
A suite is a list of checks that will run one after the other, and its results will be displayed together.
To customize a suite, we can either:
Create new custom suites, by choosing the checks (and the optional conditions) that we want the suite to have.
Modify a built-in suite by adding and/or removing checks and conditions, to adapt it to our needs.
Create a New Suite#
Let’s say we want to create our custom suite, mainly with various performance checks,
including PerformanceReport(), TrainTestDifferenceOverfit() and several more.
For assistance in understanding which checks are implemented and can be included, we suggest using any of:
Built-in suites (by printing them to see which checks they include)
from sklearn.metrics import make_scorer, precision_score, recall_score
from deepchecks.tabular import Suite
# importing all existing checks for demonstration simplicity
from deepchecks.tabular.checks import *
# The Suite's first argument is its name, and then all of the check objects.
# Some checks can receive arguments when initialized (all check arguments have default values)
# Each check can have an optional condition(/s)
# Multiple conditions can be applied subsequentially
new_custom_suite = Suite('Simple Suite For Model Performance',
ModelInfo(),
# use custom scorers for performance report:
TrainTestPerformance().add_condition_train_test_relative_degradation_less_than(threshold=0.15)\
.add_condition_test_performance_greater_than(0.8),
ConfusionMatrixReport(),
SimpleModelComparison(strategy='most_frequent',
scorers={'Recall (Multiclass)': make_scorer(recall_score, average=None),
'Precision (Multiclass)': make_scorer(precision_score, average=None)}
).add_condition_gain_greater_than(0.3)
)
# The scorers parameter can also be passed to the suite in order to override the scorers of all the checks in the suite.
# Find more about scorers at https://docs.deepchecks.com/stable/user-guide/general/metrics_guide.html.
Let’s see the suite:
Simple Suite For Model Performance: [
0: ModelInfo
1: TrainTestPerformance
Conditions:
0: Train-Test scores relative degradation is less than 0.15
1: Scores are greater than 0.8
2: ConfusionMatrixReport
3: SimpleModelComparison(alternative_scorers={'Recall (Multiclass)': make_scorer(recall_score, average=None), 'Precision (Multiclass)': make_scorer(precision_score, average=None)})
Conditions:
0: Model performance gain over simple model is greater than 30%
]
TIP: the auto-complete may not work from inside a new suite definition, so if you want to use the auto-complete to see the arguments a check receive or the built-in conditions it has, try doing it outside of the suite’s initialization.
For example, to see a check’s built-in conditions, type in a new cell: ``NameOfDesiredCheck().add_condition_`` and then check the auto-complete suggestions (using Shift + Tab), to discover the built-in checks.
Additional Notes about Conditions in a Suite#
Checks in the built-in suites come with pre-defined conditions, and when building your custom suite you should choose which conditions to add.
Most check classes have built-in methods for adding conditions. These apply to the naming convention
add_condition_..., which enables adding a condition logic to parse the check’s results.Each check instance can have several conditions or none. Each condition will be evaluated separately.
The pass (✓) / fail (✖) / insight (!) status of the conditions, along with the condition’s name and extra info will be displayed in the suite’s Conditions Summary.
Most conditions have configurable arguments that can be passed to the condition while adding it.
For more info about conditions, check out Configure a Condition.
Run the Suite#
This is simply done by calling the run() method of the suite.
To see that in action, we’ll need datasets and a model.
Let’s quickly load a dataset and train a simple model for the sake of this demo
Load Datasets and Train a Simple Model#
import numpy as np
# General imports
import pandas as pd
np.random.seed(22)
from sklearn.ensemble import RandomForestClassifier
from sklearn.model_selection import train_test_split
from deepchecks.tabular.datasets.classification import iris
# Load pre-split Datasets
train_dataset, test_dataset = iris.load_data(as_train_test=True)
label_col = 'target'
# Train Model
rf_clf = RandomForestClassifier()
rf_clf.fit(train_dataset.data[train_dataset.features],
train_dataset.data[train_dataset.label_name]);
RandomForestClassifier()
Run Suite#
new_custom_suite.run(model=rf_clf, train_dataset=train_dataset, test_dataset=test_dataset)
Simple Suite For Model Performance:
| | 0/4 [Time: 00:00]
Simple Suite For Model Performance:
|##5 | 2/4 [Time: 00:00, Check=Train Test Performance]Precision is ill-defined and being set to 0.0 in labels with no predicted samples. Use `zero_division` parameter to control this behavior.
Simple Suite For Model Performance:
|#####| 4/4 [Time: 00:00, Check=Simple Model Comparison]
Modify an Existing Suite#
from deepchecks.tabular.suites import train_test_validation
customized_suite = train_test_validation()
# let's check what it has:
customized_suite
Train Test Validation Suite: [
0: DatasetsSizeComparison
Conditions:
0: Test-Train size ratio is greater than 0.01
1: NewLabelTrainTest
Conditions:
0: Number of new label values is less or equal to 0
2: NewCategoryTrainTest
Conditions:
0: Ratio of samples with a new category is less or equal to 0%
3: StringMismatchComparison
Conditions:
0: No new variants allowed in test data
4: DateTrainTestLeakageDuplicates
Conditions:
0: Date leakage ratio is less or equal to 0%
5: DateTrainTestLeakageOverlap
Conditions:
0: Date leakage ratio is less or equal to 0%
6: IndexTrainTestLeakage
Conditions:
0: Ratio of leaking indices is less or equal to 0%
7: TrainTestSamplesMix(n_to_show=5)
Conditions:
0: Percentage of test data samples that appear in train data is less or equal to 10%
8: FeatureLabelCorrelationChange(ppscore_params={}, random_state=42)
Conditions:
0: Train-Test features' Predictive Power Score difference is less than 0.2
1: Train features' Predictive Power Score is less than 0.7
9: FeatureDrift
Conditions:
0: categorical drift score < 0.2 and numerical drift score < 0.2
10: LabelDrift
Conditions:
0: categorical drift score < 0.15 and numerical drift score < 0.15 for label drift
11: MultivariateDrift
Conditions:
0: Drift value is less than 0.25
]
# and modify it by removing a check by index:
customized_suite.remove(1)
Train Test Validation Suite: [
0: DatasetsSizeComparison
Conditions:
0: Test-Train size ratio is greater than 0.01
2: NewCategoryTrainTest
Conditions:
0: Ratio of samples with a new category is less or equal to 0%
3: StringMismatchComparison
Conditions:
0: No new variants allowed in test data
4: DateTrainTestLeakageDuplicates
Conditions:
0: Date leakage ratio is less or equal to 0%
5: DateTrainTestLeakageOverlap
Conditions:
0: Date leakage ratio is less or equal to 0%
6: IndexTrainTestLeakage
Conditions:
0: Ratio of leaking indices is less or equal to 0%
7: TrainTestSamplesMix(n_to_show=5)
Conditions:
0: Percentage of test data samples that appear in train data is less or equal to 10%
8: FeatureLabelCorrelationChange(ppscore_params={}, random_state=42)
Conditions:
0: Train-Test features' Predictive Power Score difference is less than 0.2
1: Train features' Predictive Power Score is less than 0.7
9: FeatureDrift
Conditions:
0: categorical drift score < 0.2 and numerical drift score < 0.2
10: LabelDrift
Conditions:
0: categorical drift score < 0.15 and numerical drift score < 0.15 for label drift
11: MultivariateDrift
Conditions:
0: Drift value is less than 0.25
]
from deepchecks.tabular.checks import UnusedFeatures
# and add a new check with a condition:
customized_suite.add(
UnusedFeatures().add_condition_number_of_high_variance_unused_features_less_or_equal())
Train Test Validation Suite: [
0: DatasetsSizeComparison
Conditions:
0: Test-Train size ratio is greater than 0.01
2: NewCategoryTrainTest
Conditions:
0: Ratio of samples with a new category is less or equal to 0%
3: StringMismatchComparison
Conditions:
0: No new variants allowed in test data
4: DateTrainTestLeakageDuplicates
Conditions:
0: Date leakage ratio is less or equal to 0%
5: DateTrainTestLeakageOverlap
Conditions:
0: Date leakage ratio is less or equal to 0%
6: IndexTrainTestLeakage
Conditions:
0: Ratio of leaking indices is less or equal to 0%
7: TrainTestSamplesMix(n_to_show=5)
Conditions:
0: Percentage of test data samples that appear in train data is less or equal to 10%
8: FeatureLabelCorrelationChange(ppscore_params={}, random_state=42)
Conditions:
0: Train-Test features' Predictive Power Score difference is less than 0.2
1: Train features' Predictive Power Score is less than 0.7
9: FeatureDrift
Conditions:
0: categorical drift score < 0.2 and numerical drift score < 0.2
10: LabelDrift
Conditions:
0: categorical drift score < 0.15 and numerical drift score < 0.15 for label drift
11: MultivariateDrift
Conditions:
0: Drift value is less than 0.25
12: UnusedFeatures
Conditions:
0: Number of high variance unused features is less or equal to 5
]
# lets remove all condition for the FeatureLabelCorrelationChange:
customized_suite[3].clean_conditions()
# and update the suite's name:
customized_suite.name = 'New Data Leakage Suite'
# and now we can run our modified suite:
customized_suite.run(train_dataset, test_dataset, rf_clf)
New Data Leakage Suite:
| | 0/12 [Time: 00:00]
New Data Leakage Suite:
|######## | 8/12 [Time: 00:00, Check=Feature Label Correlation Change]
Total running time of the script: ( 0 minutes 2.730 seconds)